Approximate Bit Vectors for Fast Unification

Bit vectors provide a way to compute the existence of least upper bounds in partial orders, which is a fundamental operation needed by any unificationbased parser. However, bit vectors have seen relatively little adoption because of their length and associated speed disadvantages. We present a novel bit vector technique based on allowing one-sided errors; the resulting approximate bit vectors can be much shorter than the minimum lengths required by existing techniques that would provide exact answers. We give experimental results showing that our approximate vectors give accurate enough answers to be useful in practice.